This project implements a deep learning solution for image super-resolution, specifically upscaling 16x16 face images by a 4x factor to 64x64. The architecture is a DCGAN where the generator takes low-resolution input images rather than random noise, combining an adversarial loss with L1 reconstruction loss to ensure both plausibility and fidelity to the input. The generator employs ResNet modules for faster training convergence. The model was trained on a GTX 1080 GPU for approximately 10 epochs.